Reordered local vertices in load_mesh / flipped tets in mesh

[Huangzizhou]

When a mesh in msh22 format is loaded, if I call mesh->cell_vertex(c, v) to get the cell list, the local vertices are sorted.

Here're the first few cells in my mesh

$Elements
31443
1 4 2 0 0 4555 6346 6399 1253
2 4 2 0 0 6346 5707 6399 1253
3 4 2 0 0 6399 1262 4555 1253
4 4 2 0 0 5707 1262 6399 1253
5 4 2 0 0 3491 6716 228 233

Here're the first few cells in the mesh class

1252 4554 6345 6398 
1252 5706 6345 6398 
1252 1261 4554 6398 
1252 1261 5706 6398 
227 232 3490 6715 

This may make tets flip if one wants to export the mesh.

[teseoch]

this should fix this one too:
68defd5

[zfergus]

I don't think this issue was ever properly resolved. The commit linked does not fix the problem which is that the CMesh3D class sorts each of the cell's vertices by ID.

In CMesh3D::load(const GEO::Mesh &M), there is a sort of the cell's vertices:

for (auto fid : cell.fs)
{
    cell.vs.insert(cell.vs.end(), mesh_.faces[fid].vs.begin(), mesh_.faces[fid].vs.end());
}
sort(cell.vs.begin(), cell.vs.end());
cell.vs.erase(unique(cell.vs.begin(), cell.vs.end()), cell.vs.end());

Maybe this is only so the unique will work, and removing duplicates while preserving order might fix it. 🤞

There are a number of other sorts in that file as well. Some are doing it to remove duplicates, but others are doing it on temporary local variables to check for equality. The latter seems perfectly fine.

[zfergus]

Actually even with an order-preserving duplicate removal, I still get inverted elements. I think it comes from the for-loop. It looks like it is creating a cell out of the cell's faces rather than using the cell that was given.

[Huangzizhou]

Yes, I once went through this, I also think it's creating every cell from its faces.